def griffin_lim(mag, phase_angle, n_fft, hop, num_iters):
"""Iterative algorithm for phase retrival from a magnitude spectrogram.
Args:
mag: Magnitude spectrogram.
phase_angle: Initial condition for phase.
n_fft: Size of the FFT.
hop: Stride of FFT. Defaults to n_fft/2.
num_iters: Griffin-Lim iterations to perform.
Returns:
audio: 1-D array of float32 sound samples.
"""
fft_config = dict(n_fft=n_fft, win_length=n_fft, hop_length=hop, center=True)
ifft_config = dict(win_length=n_fft, hop_length=hop, center=True)
complex_specgram = inv_magphase(mag, phase_angle)
for i in range(num_iters):
audio = librosa.istft(complex_specgram, **ifft_config)
if i != num_iters - 1:
complex_specgram = librosa.stft(audio, **fft_config)
_, phase = librosa.magphase(complex_specgram)
phase_angle = np.angle(phase)
complex_specgram = inv_magphase(mag, phase_angle)
return audio
python类istft()的实例源码
def test_stft_istft(self):
try:
import librosa
ds = F.load_digit_wav()
name = ds.keys()[0]
path = ds[name]
y, _ = speech.read(path, pcm=True)
hop_length = int(0.01 * 8000)
stft = signal.stft(y, n_fft=256, hop_length=hop_length, window='hann')
stft_ = librosa.stft(y, n_fft=256, hop_length=hop_length, window='hann')
self.assertTrue(np.allclose(stft, stft_.T))
y1 = signal.istft(stft, hop_length=hop_length, window='hann')
y2 = librosa.istft(stft_, hop_length=hop_length, window='hann')
self.assertTrue(np.allclose(y1, y2))
except ImportError:
print("test_stft_istft require librosa.")
def NMF(stft, n_sources):
"""
Sound source separation using NMF
:param stft: the short-time Fourier Transform of the signal
:param n_sources: the number of sources
:returns:
- Ys: sources
"""
print "Computing approximations"
W, H = librosa.decompose.decompose(np.abs(stft), n_components=n_sources, sort=True)
print "Reconstructing signals"
Ys = list(scipy.outer(W[:,i], H[i])*np.exp(1j*np.angle(stft)) for i in xrange(n_sources))
print "Saving sound arrays"
ys = [librosa.istft(Y) for Y in Ys]
del Ys
return ys
def istft_mc(X,hop,dtype=np.float32,nsampl=None,flag_noDiv=0,window=None):
#assumes X is of shape F x nfram x nch, where F=Nwin/2+1
#returns xr of shape nch x nsampl
N=2*(X.shape[0]-1)
nch=X.shape[2]
for ich in range(0,nch):
X0=X[:,:,ich]
if flag_noDiv:
x0r=istft_noDiv(X0,hop_length=hop,center=False,window=window,dtype=dtype)
else:
x0r=librosa.core.istft(X0,hop_length=hop,center=False,window=window,dtype=dtype)
if ich==0:
xr=np.zeros((nch,len(x0r))).astype(dtype)
xr[0,:]=x0r
else:
xr[ich,:]=x0r
#trim off extra zeros
nfram=xr.shape[1]
xr=xr[:,0:(nfram-N)]
nfram=xr.shape[1]
xr=xr[:,N:]
if not nsampl is None:
xr=xr[:,0:nsampl]
return xr, N
def griffinlim(spectrogram, n_iter=50, window='hann', n_fft=2048, win_length=2048, hop_length=-1, verbose=False):
if hop_length == -1:
hop_length = n_fft // 4
angles = np.exp(2j * np.pi * np.random.rand(*spectrogram.shape))
t = tqdm(range(n_iter), ncols=100, mininterval=2.0, disable=not verbose)
for i in t:
full = np.abs(spectrogram).astype(np.complex) * angles
inverse = librosa.istft(full, hop_length = hop_length, win_length = win_length, window = window)
rebuilt = librosa.stft(inverse, n_fft = n_fft, hop_length = hop_length, win_length = win_length, window = window)
angles = np.exp(1j * np.angle(rebuilt))
if verbose:
diff = np.abs(spectrogram) - np.abs(rebuilt)
t.set_postfix(loss=np.linalg.norm(diff, 'fro'))
full = np.abs(spectrogram).astype(np.complex) * angles
inverse = librosa.istft(full, hop_length = hop_length, win_length = win_length, window = window)
return inverse
def reconstructed_audio_from_spectrogram(self) -> ndarray:
return librosa.istft(self._complex_spectrogram(), win_length=self.fourier_window_length,
hop_length=self.hop_length)
def _griffin_lim(S, n_fft, win_length, hop_length, num_iters):
angles = np.exp(2j * np.pi * np.random.rand(*S.shape))
S_complex = np.abs(S).astype(np.complex)
for i in range(num_iters):
if i > 0:
angles = np.exp(1j * np.angle(librosa.stft(y=y, n_fft=n_fft, hop_length=hop_length, win_length=win_length)))
y = librosa.istft(S_complex * angles, hop_length=hop_length, win_length=win_length)
return y
def create_audio_from_spectrogram(spec):
spec_transposed = tf.transpose(spec).eval()
return librosa.istft(spec_transposed, Config.hop_length)
def invert_spectrogram(spectrogram):
'''
spectrogram: [f, t]
'''
return librosa.istft(spectrogram, hp.hop_length, win_length=hp.win_length, window="hann")
def recons_spec_phase(Sxx_r, phase):
Sxx_r = np.sqrt(10**Sxx_r)
R = np.multiply(Sxx_r , phase)
result = librosa.istft(R,
hop_length=256,
win_length=512,
window=scipy.signal.hamming)
return result
def expand(self, audio):
ori_len = audio.shape[0]
tmp = resample(audio, r=0.5, type='sinc_best')
down_len = tmp.shape[0]
tmp = resample(tmp, r=(ori_len+1) / float(down_len), type='sinc_best')
tmp = librosa.stft(audio, 1024)
phase = np.divide(tmp, np.abs(tmp))
spec_input = np.abs(librosa.stft(audio, 1024))[0:n_input, ::]
spec_input = spec_input[::, 0:spec_input.shape[1]//n_len*n_len]
spec_input = np.split(spec_input,
spec_input.shape[1]//n_len, axis=1)
spec_input = np.asarray(spec_input)
spec_input = np.expand_dims(spec_input, axis=-1)
feed_dict = {self.input_op: np.log1p(spec_input) / 12.0}
debug = self.sess.run(self.debug_op, feed_dict=feed_dict)
np.save('debug.npy', debug)
S = self.sess.run(self.eva_op, feed_dict=feed_dict)
S[S >= 5e3] = 5e3
S[S <= 0] = 0
print ('mean', np.mean(S))
print (np.sum(np.isinf(S)))
S = np.squeeze(np.concatenate(np.split(S, S.shape[0]), axis=2),
axis=(0, -1))
phase = phase[..., :S.shape[1]]
print (phase.shape)
print (S.shape)
print (np.sum(np.isinf(np.multiply(S, phase))))
X = librosa.istft(np.multiply(S, phase))
return X
def mag2spec_and_audio_wiener(xhat, recons, MS, MSphase, arguments):
#xhat = xhat.cpu().numpy()
#recons = recons.cpu().numpy()
try: # pytorch case
MS = MS.cpu().numpy()
MSphase = MSphase.cpu().numpy()
Nmix = MSphase.shape[0]
maghats = np.split(xhat, Nmix, axis=0)
reconss = np.split(recons, Nmix, axis=0)
mixmags = np.split(MS, Nmix, axis=0)
phases = np.split(MSphase, Nmix, axis=0)
except:
maghats = [xhat]
reconss = [recons]
mixmags = [MS]
phases = [MSphase]
all_audio = []
eps = 1e-20
for maghat, recons, mixmag, phase in zip(maghats, reconss, mixmags, phases):
mask = (maghat / (recons + eps))
all_audio.append(lr.istft((mask*mixmag*np.exp(1j*phase)).transpose(),
win_length=arguments.win_length))
return all_audio, maghats
def mag2spec_and_audio(xhat, MSphase, arguments):
MSphase = MSphase.cpu().numpy()
Nmix = MSphase.shape[0]
mags = np.split(xhat, Nmix, axis=0)
phases = np.split(MSphase, Nmix, axis=0)
all_audio = []
for mag, phase in zip(mags, phases):
all_audio.append(lr.istft((mag*np.exp(1j*phase.squeeze())).transpose(),
win_length=arguments.win_length))
return all_audio, mags
def invert_spectrogram(spectrogram):
'''
spectrogram: [f, t]
'''
return librosa.istft(spectrogram, hp.hop_length, win_length=hp.win_length, window="hann")
def relaxed_music_remix(self, neg_arous_dir, files, decisions):
neg_arousal_h = self.sad_music_remix(neg_arous_dir, files, decisions, harmonic = True)
relaxed_harmonic = librosa.istft(neg_arousal_h)
onsets = hfc_onsets(np.float32(relaxed_harmonic))
interv = seconds_to_indices(onsets)
steps = overlapped_intervals(interv)
output = librosa.effects.remix(relaxed_harmonic, steps[::-1], align_zeros = True)
output = librosa.effects.pitch_shift(output, sr = 44100, n_steps = 4)
remix_filename = 'data/emotions/remixes/relaxed/'+str(time.strftime("%Y%m%d-%H:%M:%S"))+'multitag_remix.ogg'
MonoWriter(filename=remix_filename, format = 'ogg', sampleRate = 44100)(output)
subprocess.call(["ffplay", "-nodisp", "-autoexit", remix_filename])
def angry_music_remix(self, pos_arous_dir, files, decisions):
pos_arousal_h = self.happy_music_remix(pos_arous_dir, files, decisions, harmonic = True)
angry_harmonic = librosa.istft(pos_arousal_h)
interv = RhythmExtractor2013()(angry_harmonic)[1] * 44100
steps = overlapped_intervals(interv)
output = librosa.effects.remix(angry_harmonic, steps, align_zeros = True)
output = librosa.effects.pitch_shift(output, sr = 44100, n_steps = 3)
remix_filename = 'data/emotions/remixes/angry/'+str(time.strftime("%Y%m%d-%H:%M:%S"))+'multitag_remix.ogg'
MonoWriter(filename=remix_filename, format = 'ogg', sampleRate = 44100)(np.float32(output))
subprocess.call(["ffplay", "-nodisp", "-autoexit", remix_filename])
def _griffin_lim_tensorflow(S):
'''TensorFlow implementation of Griffin-Lim
Based on https://github.com/Kyubyong/tensorflow-exercises/blob/master/Audio_Processing.ipynb
'''
with tf.variable_scope('griffinlim'):
# TensorFlow's stft and istft operate on a batch of spectrograms; create batch of size 1
S = tf.expand_dims(S, 0)
S_complex = tf.identity(tf.cast(S, dtype=tf.complex64))
y = _istft_tensorflow(S_complex)
for i in range(hparams.griffin_lim_iters):
est = _stft_tensorflow(y)
angles = est / tf.cast(tf.maximum(1e-8, tf.abs(est)), tf.complex64)
y = _istft_tensorflow(S_complex * angles)
return tf.squeeze(y, 0)
def _istft(y):
_, hop_length, win_length = _stft_parameters()
return librosa.istft(y, hop_length=hop_length, win_length=win_length)
def ispecgram(spec,
n_fft=512,
hop_length=None,
mask=True,
log_mag=True,
re_im=False,
dphase=True,
mag_only=True,
num_iters=1000):
"""Inverse Spectrogram using librosa.
Args:
spec: 3-D specgram array [freqs, time, (mag_db, dphase)].
n_fft: Size of the FFT.
hop_length: Stride of FFT. Defaults to n_fft/2.
mask: Reverse the mask of the phase derivative by the magnitude.
log_mag: Use the logamplitude.
re_im: Output Real and Imag. instead of logMag and dPhase.
dphase: Use derivative of phase instead of phase.
mag_only: Specgram contains no phase.
num_iters: Number of griffin-lim iterations for mag_only.
Returns:
audio: 1-D array of sound samples. Peak normalized to 1.
"""
if not hop_length:
hop_length = n_fft // 2
ifft_config = dict(win_length=n_fft, hop_length=hop_length, center=True)
if mag_only:
mag = spec[:, :, 0]
phase_angle = np.pi * np.random.rand(*mag.shape)
elif re_im:
spec_real = spec[:, :, 0] + 1.j * spec[:, :, 1]
else:
mag, p = spec[:, :, 0], spec[:, :, 1]
if mask and log_mag:
p /= (mag + 1e-13 * np.random.randn(*mag.shape))
if dphase:
# Roll up phase
phase_angle = np.cumsum(p * np.pi, axis=1)
else:
phase_angle = p * np.pi
# Magnitudes
if log_mag:
mag = (mag - 1.0) * 120.0
mag = 10**(mag / 20.0)
phase = np.cos(phase_angle) + 1.j * np.sin(phase_angle)
spec_real = mag * phase
if mag_only:
audio = griffin_lim(
mag, phase_angle, n_fft, hop_length, num_iters=num_iters)
else:
audio = librosa.core.istft(spec_real, **ifft_config)
return np.squeeze(audio / audio.max())
def bpm_cluster_files(files_dir, descriptor, euclidean_labels, files):
"""
locate bpm according to clusters in clusters directories
:param files_dir: directory where sounds are located
:param descriptor: descriptor used for similarity
:param euclidean_labels: groups of clusters
:param files: the .json files (use get_files)
"""
groups = [[] for i in xrange(len(np.unique(euclidean_labels)))]
for i, x in enumerate(euclidean_labels):
groups[x].append([files[0][i], x])
for c in xrange(len(groups)):
if not os.path.exists(files_dir+'/tempo/'+str(c)):
os.makedirs(files_dir+'/tempo/'+str(c))
os.makedirs(files_dir+'/tempo/'+str(c)+'/remix')
for t in groups[c]:
for s in list(os.walk(files_dir+'/tempo', topdown=False))[-1][-1]:
if str(t[0]).split('.json')[0] == s.split('tempo.ogg')[0]:
shutil.copy(files_dir+'/tempo/'+s, files_dir+'/tempo/'+str(c)+'/'+s)
print str().join((str(t)))
try:
simil_audio = [MonoLoader(filename=files_dir+'/tempo/'+str(c)+f)() for f in list(os.walk(files_dir+'/tempo/'+str(c), topdown = False))[-1][-1]]
audio0 = scratch_music(choice(simil_audio))
audio1 = scratch_music(choice(simil_audio))
del simil_audio
audio_N = min([len(i) for i in [audio0, audio1]])
audio_samples = [i[:audio_N]/i.max() for i in [audio0, audio1]]
simil_x = np.array(audio_samples).sum(axis=0)
del audio_samples
simil_x = 0.5*simil_x/simil_x.max()
h, p = librosa.decompose.hpss(librosa.core.stft(simil_x))
del simil_x, h
p = librosa.istft(p)
MonoWriter(filename=files_dir+'/tempo/'+str(c)+'/remix/'+'similarity_mix_bpm.ogg', format = 'ogg', sampleRate = 44100)(p)
del p
except Exception, e:
print logger.exception(e)
continue
# save mfcc clusters files in mfcc clusters directory
def mfcc_cluster_files(files_dir, descriptor, euclidean_labels, files):
"""
locate mfcc files according to clusters in clusters directories
:param files_dir: directory where sounds are located
:param descriptor: descriptor used for similarity
:param euclidean_labels: groups of clusters
:param files: the .json files (use get_files)
"""
groups = [[] for i in xrange(len(np.unique(euclidean_labels)))]
for i, x in enumerate(euclidean_labels):
groups[x].append([files[0][i], x])
for c in xrange(len(groups)):
if not os.path.exists(files_dir+'/mfcc/'+str(c)):
os.makedirs(files_dir+'/mfcc/'+str(c))
os.makedirs(files_dir+'/mfcc/'+str(c)+'/remix')
for t in groups[c]:
for s in list(os.walk(files_dir+'/mfcc', topdown=False))[-1][-1]:
if str(t[0]).split('.')[0] == s.split('mfcc.ogg')[0]:
shutil.copy(files_dir+'/mfcc/'+s, files_dir+'/mfcc/'+str(c)+'/'+s)
print t
try:
simil_audio = [MonoLoader(filename=files_dir+'/mfcc/'+str(c)+f)() for f in list(os.walk(files_dir+'/mfcc/'+str(c), topdown = False))[-1][-1]]
audio0 = scratch_music(choice(simil_audio))
audio1 = scratch_music(choice(simil_audio))
del simil_audio
audio_N = min([len(i) for i in [audio0, audio1]])
audio_samples = [i[:audio_N]/i.max() for i in [audio0, audio1]]
simil_x = np.array(audio_samples).sum(axis=0)
del audio_samples
simil_x = 0.5*simil_x/simil_x.max()
h, p = librosa.decompose.hpss(librosa.core.stft(simil_x))
del simil_x, p
h = librosa.istft(h)
MonoWriter(filename=files_dir+'/mfcc/'+str(c)+'/remix/'+'similarity_mix_mfcc.ogg', format = 'ogg', sampleRate = 44100)(h)
del h
except Exception, e:
print e
continue
# save contrast cluster files in contrast cluster directory
def contrast_cluster_files(files_dir, descriptor, euclidean_labels, files):
"""
locate contrast files according to clusters in clusters directories
:param files_dir: directory where sounds are located
:param descriptor: descriptor used for similarity
:param euclidean_labels: groups of clusters
:param files: the .json files (use get_files)
"""
groups = [[] for i in xrange(len(np.unique(euclidean_labels)))]
for i, x in enumerate(euclidean_labels):
groups[x].append([files[0][i], x])
for c in xrange(len(groups)):
if not os.path.exists(files_dir+'/valleys/'+str(c)):
os.makedirs(files_dir+'/valleys/'+str(c))
os.makedirs(files_dir+'/valleys/'+str(c)+'/remix')
for t in groups[c]:
for s in list(os.walk(files_dir+'/valleys', topdown=False))[-1][-1]:
if str(t[0]).split('.')[0] == s.split('valleys.ogg')[0]:
shutil.copy(files_dir+'/valleys/'+s, files_dir+'/valleys/'+str(c)+'/'+s)
print t
try:
simil_audio = [MonoLoader(filename=files_dir+'/valleys/'+str(c)+f)() for f in list(os.walk(files_dir+'/valleys/'+str(c), topdown = False))[-1][-1]]
audio0 = scratch_music(choice(simil_audio))
audio1 = scratch_music(choice(simil_audio))
del simil_audio
audio_N = min([len(i) for i in [audio0, audio1]])
audio_samples = [i[:audio_N]/i.max() for i in [audio0, audio1]]
simil_x = np.array(audio_samples).sum(axis=0)
del audio_samples
simil_x = 0.5*simil_x/simil_x.max()
h, p = librosa.decompose.hpss(librosa.core.stft(simil_x))
del simil_x, p
h = librosa.istft(h)
MonoWriter(filename=files_dir+'/valleys/'+str(c)+'/remix/'+'similarity_mix_valleys.ogg', format = 'ogg', sampleRate = 44100)(h)
del h
except Exception, e:
print e
continue
# save centroid cluster files in centroid cluster directory
def loudness_cluster_files(files_dir, descriptor, euclidean_labels, files):
"""
locate loudness files according to clusters in clusters directories
:param files_dir: directory where sounds are located
:param descriptor: descriptor used for similarity
:param euclidean_labels: groups of clusters
:param files: the .json files (use get_files)
"""
groups = [[] for i in xrange(len(np.unique(euclidean_labels)))]
for i, x in enumerate(euclidean_labels):
groups[x].append([files[0][i], x])
for c in xrange(len(groups)):
if not os.path.exists(files_dir+'/loudness/'+str(c)):
os.makedirs(files_dir+'/loudness/'+str(c))
os.makedirs(files_dir+'/loudness/'+str(c)+'/remix')
for t in groups[c]:
for s in list(os.walk(files_dir+'/loudness', topdown=False))[-1][-1]:
if str(t[0]).split('.')[0] == s.split('loudness.ogg')[0]:
shutil.copy(files_dir+'/loudness/'+s, files_dir+'/loudness/'+str(c)+'/'+s)
print t
try:
simil_audio = [MonoLoader(filename=files_dir+'/loudness/'+str(c)+f)() for f in list(os.walk(files_dir+'/loudness/'+str(c), topdown = False))[-1][-1]]
audio0 = scratch_music(choice(simil_audio))
audio1 = scratch_music(choice(simil_audio))
del simil_audio
audio_N = min([len(i) for i in [audio0, audio1]])
audio_samples = [i[:audio_N]/i.max() for i in [audio0, audio1]]
simil_x = np.array(audio_samples).sum(axis=0)
del audio_samples
simil_x = 0.5*simil_x/simil_x.max()
h, p = librosa.decompose.hpss(librosa.core.stft(simil_x))
del simil_x, p
h = librosa.istft(h)
MonoWriter(filename=files_dir+'/loudness/'+str(c)+'/remix/'+'similarity_mix_centroid.ogg', format = 'ogg', sampleRate = 44100)(h)
del h
except Exception, e:
print e
continue
# save hfc cluster files in hfc cluster directory
def hfc_cluster_files(files_dir, descriptor, euclidean_labels, files):
"""
locate hfc files according to clusters in clusters directories
:param files_dir: directory where sounds are located
:param descriptor: descriptor used for similarity
:param euclidean_labels: groups of clusters
:param files: the .json files (use get_files)
"""
groups = [[] for i in xrange(len(np.unique(euclidean_labels)))]
for i, x in enumerate(euclidean_labels):
groups[x].append([files[0][i], x])
for c in xrange(len(groups)):
if not os.path.exists(files_dir+'/hfc/'+str(c)):
os.makedirs(files_dir+'/hfc/'+str(c))
os.makedirs(files_dir+'/hfc/'+str(c)+'/remix')
for t in groups[c]:
for s in list(os.walk(files_dir+'/hfc', topdown=False))[-1][-1]:
if str(t[0]).split('.')[0] == s.split('hfc.ogg')[0]:
shutil.copy(files_dir+'/hfc/'+s, files_dir+'/hfc/'+str(c)+'/'+s)
print t
try:
simil_audio = [MonoLoader(filename=files_dir+'/hfc/'+str(c)+f)() for f in list(os.walk(files_dir+'/hfc/'+str(c), topdown = False))[-1][-1]]
audio0 = scratch_music(choice(simil_audio))
audio1 = scratch_music(choice(simil_audio))
del simil_audio
audio_N = min([len(i) for i in [audio0, audio1]])
audio_samples = [i[:audio_N]/i.max() for i in [audio0, audio1]]
simil_x = np.array(audio_samples).sum(axis=0)
del audio_samples
simil_x = 0.5*simil_x/simil_x.max()
h, p = librosa.decompose.hpss(librosa.core.stft(simil_x))
del simil_x, h
p = librosa.istft(p)
MonoWriter(filename=files_dir+'/hfc/'+str(c)+'/remix/'+'similarity_mix_hfc.ogg', format = 'ogg', sampleRate = 44100)(p)
del p
except Exception, e:
print e
continue
# save inharmonicity cluster files in inharmonicity cluster directory
def inharmonicity_cluster_files(files_dir, descriptor, euclidean_labels, files):
"""
locate inharmonicity files according to clusters in clusters directories
:param files_dir: directory where sounds are located
:param descriptor: descriptor used for similarity
:param euclidean_labels: groups of clusters
:param files: the .json files (use get_files)
"""
groups = [[] for i in xrange(len(np.unique(euclidean_labels)))]
for i, x in enumerate(euclidean_labels):
groups[x].append([files[0][i], x])
for c in xrange(len(groups)):
if not os.path.exists(files_dir+'/inharmonicity/'+str(c)):
os.makedirs(files_dir+'/inharmonicity/'+str(c))
os.makedirs(files_dir+'/inharmonicity/'+str(c)+'/remix')
for t in groups[c]:
for s in list(os.walk(files_dir+'/inharmonicity', topdown=False))[-1][-1]:
if str(t[0]).split('.')[0] == s.split('inharmonicity.ogg')[0]:
shutil.copy(files_dir+'/inharmonicity/'+s, files_dir+'/inharmonicity/'+str(c)+'/'+s)
print t
try:
simil_audio = [MonoLoader(filename=files_dir+'/inharmonicity/'+str(c)+f)() for f in list(os.walk(files_dir+'/inharmonicity/'+str(c), topdown = False))[-1][-1]]
audio0 = scratch_music(choice(simil_audio))
audio1 = scratch_music(choice(simil_audio))
del simil_audio
audio_N = min([len(i) for i in [audio0, audio1]])
audio_samples = [i[:audio_N]/i.max() for i in [audio0, audio1]]
simil_x = np.array(audio_samples).sum(axis=0)
del audio_samples
simil_x = 0.5*simil_x/simil_x.max()
h, p = librosa.decompose.hpss(librosa.core.stft(simil_x))
del simil_x, p
h = librosa.istft(h)
MonoWriter(filename=files_dir+'/inharmonicity/'+str(c)+'/remix/'+'similarity_mix_inharmonicity.ogg', format = 'ogg', sampleRate = 44100)(h)
del h
except Exception, e:
print e
continue
# save dissonance cluster files in dissonance cluster directory
def dissonance_cluster_files(files_dir, descriptor, euclidean_labels, files):
"""
locate dissonance files according to clusters in clusters directories
:param files_dir: directory where sounds are located
:param descriptor: descriptor used for similarity
:param euclidean_labels: groups of clusters
:param files: the .json files (use get_files)
"""
groups = [[] for i in xrange(len(np.unique(euclidean_labels)))]
for i, x in enumerate(euclidean_labels):
groups[x].append([files[0][i], x])
for c in xrange(len(groups)):
if not os.path.exists(files_dir+'/dissonance/'+str(c)):
os.makedirs(files_dir+'/dissonance/'+str(c))
os.makedirs(files_dir+'/dissonance/'+str(c)+'/remix')
for t in groups[c]:
for s in list(os.walk(files_dir+'/dissonance', topdown=False))[-1][-1]:
if str(t[0]).split('.')[0] == s.split('dissonance.ogg')[0]:
shutil.copy(files_dir+'/dissonance/'+s, files_dir+'/dissonance/'+str(c)+'/'+s)
print t
try:
simil_audio = [MonoLoader(filename=files_dir+'/dissonance/'+str(c)+f)() for f in list(os.walk(files_dir+'/dissonance/'+str(c), topdown = False))[-1][-1]]
audio0 = scratch_music(choice(simil_audio))
audio1 = scratch_music(choice(simil_audio))
del simil_audio
audio_N = min([len(i) for i in [audio0, audio1]])
audio_samples = [i[:audio_N]/i.max() for i in [audio0, audio1]]
simil_x = np.array(audio_samples).sum(axis=0)
del audio_samples
simil_x = 0.5*simil_x/simil_x.max()
h, p = librosa.decompose.hpss(librosa.core.stft(simil_x))
del simil_x, p
h = librosa.istft(h)
MonoWriter(filename=files_dir+'/dissonance/'+str(c)+'/remix/'+'similarity_mix_dissonance.ogg', format = 'ogg', sampleRate = 44100)(h)
del h
except Exception, e:
print e
continue
# save attack cluster files in attack cluster directory
def attack_cluster_files(files_dir, descriptor, euclidean_labels, files):
"""
locate attack files according to clusters in clusters directories
:param files_dir: directory where sounds are located
:param descriptor: descriptor used for similarity
:param euclidean_labels: groups of clusters
:param files: the .json files (use get_files)
"""
groups = [[] for i in xrange(len(np.unique(euclidean_labels)))]
for i, x in enumerate(euclidean_labels):
groups[x].append([files[0][i], x])
for c in xrange(len(groups)):
if not os.path.exists(files_dir+'/attack/'+str(c)):
os.makedirs(files_dir+'/attack/'+str(c))
os.makedirs(files_dir+'/attack/'+str(c)+'/remix')
for t in groups[c]:
for s in list(os.walk(files_dir+'/attack', topdown=False))[-1][-1]:
if str(t[0]).split('.')[0] == s.split('attack.ogg')[0]:
shutil.copy(files_dir+'/attack/'+s, files_dir+'/attack/'+str(c)+'/'+s)
print t
try:
simil_audio = [MonoLoader(filename=files_dir+'/attack/'+str(c)+f)() for f in list(os.walk(files_dir+'/attack/'+str(c), topdown = False))[-1][-1]]
audio0 = scratch_music(choice(simil_audio))
audio1 = scratch_music(choice(simil_audio))
del simil_audio
audio_N = min([len(i) for i in [audio0, audio1]])
audio_samples = [i[:audio_N]/i.max() for i in [audio0, audio1]]
simil_x = np.array(audio_samples).sum(axis=0)
del audio_samples
simil_x = 0.5*simil_x/simil_x.max()
h, p = librosa.decompose.hpss(librosa.core.stft(simil_x))
del simil_x, h
p = librosa.istft(p)
MonoWriter(filename=files_dir+'/attack/'+str(c)+'/remix/'+'similarity_mix_attack.ogg', format = 'ogg', sampleRate = 44100)(p)
del p
except Exception, e:
print e
continue
# save duration cluster files in dissonance cluster directory
